No, The US Did Not Ban All Nvidia Chips from China
Here's what the US did restrict and how it's affecting Nvidia and AMD
The US Commerce Department put in place new rules on August 26th, that require a license in order to export high end machine learning chips that meet or exceed a certain performance threshold, to Russia, Hong Kong or China. The rules are meant to address the risk that such chips might have a ‘military end use’ or ‘military end user’ in China or Russia.
The US has not made the details of the restrictions public, but Nvidia mentioned them when it filed an update to its earnings projections with the US SEC. The licensing requirements apply to Nvidia's two-year-old A100 and forthcoming H100 GPUs and AMD's MI250 accelerator chips. These chips are used in data centers to train machine learning algorithms. For example, the Verge notes that Meta uses a few thousand A100 chips in its AI Research Supercluster.
In its filing, Nvidia says that about $400 million of its expected $5.9 billion in sales in Q3 "may be subject to the new license requirement." Nvidia said it may seek a license to sell the chips in some cases, and Thursday it received authorization to export A100s through March 1st in support of US customers. It also got approvals to make sure its development of the H100 is not affected, as it does some research and development in China. Nvidia no longer sells chips in Russia.
The Wall Street Journal notes that Alibaba, which operates China's largest cloud service and Tencent which also sells cloud services and is one of China's largest video and social media companies-- both are big customers of Nvidia and have bought A100s in the past. It's unclear how much of their business involves the two restricted Nvidia chips.
Chinese companies may change suppliers to companies located in Europe and Israel.